I had kinda the opposite experience using Bard. It made up random technical details for the questions I asked. It assured me that it knew the contents of a document that it didn’t have access to and generated what it claimed was information from that document rather than saying it couldn’t read it and didn’t know.
Comment on Microsoft now has implemented "compare with Bing chat" button when you visit Google Bard in Edge
sndrtj@feddit.nl 1 year ago
While bard got some really bad reviews compared to chatgpt, I’ve honestly found it better with queries chatgpt struggled with. I have two examples:
- “Please write a python function that returns an exact solution to pi” Both come up with gregory-leibnitz formula, which of course is an approximation. When challenged that I want an exact solution instead of an approximation, chatgpt apologizes and then returns yet another approximation. Bard correctly claims that that is impossible as pi is an irrational number.
- “What can you tell me about a compound called polysac-active in cough syrups”. Chatgpt hallucinates something about a company in Indonesia, which seems to have a product that sounds vaguely similar to one of the brands selling that compound. Bard, on the other hand, correctly surmises it’s mostly honey and even gives some examples of real products that feature this ingredient.
In general, Bard in general seems one of the few llms that will tell me it doesn’t know something or that something is impossible. IMHO that’s better than just coming up with a hallucination.
magnusrufus@lemmy.world 1 year ago
Rakn@discuss.tchncs.de 1 year ago
Same. Found it to be better with some and worse work others. With some of those where it provides worse results compared to ChatGPT it just feels like it’s missing the fine tuning. It provides pretty similar results as when ChatGPT 3.5 came out a while ago. People just tend to forget about it.